allthatwhichis's idea is pretty nice, although I think that the small dimension of the beam is the fast axis. I was also under the assumption that a waveplate simply rotated the polarization of the beam, not the pattern of the beam stripe itself.
I had similar ideas of just chopping the long dimension of the beam in half and recombining into a 2mm square beam instead of a 4mm by 2mm beam, but mccarrot reminded me that the small dimension of the beam was the fast axis. Therefore I would have been cutting the slow axis in half, resulting in a beam that would be an even thinner stripe at long distances.
So I'm thinking that a knife-edged setup will provide the best beam specs as you would have your fast axis on a single axis of space whereas with a PBS cube combined setup you're rotating the fast axis of one diode 90-degrees, therefore placing the high divergence of that fast axis in both the horizontal and vertical directions.
I can't remember the typical fast axis to slow axis divergence ratio for these diodes, but say that you have a fast axis divergence of 2.5mRad and a slow axis divergence of 0.5mRad. With a knife edge setup, you'll retain the same divergence figures as the two diodes' fast axes are parallel to each other.
Therefore you'd have a divergence of 2.5mRad by 0.5mRad with a knife-edged setup or a 2.5mRad by 2.5mRad divergence with a PBS setup.
Although I've never seen either in person, on paper the squarer beam of a knife-edged setup looks better than the cross shape beam, in my opinion. As the beams individually diverge they will converge together into a solid rectangular/square beam. With a PBS cube setup, both axes diverge at the same rate so you're always stuck with the cross shape.
Furthermore you'll loose more power by using a PBS cube if you use high quality mirrors.
For more fun you could take two knife-edged setups and combine them using a PBS cube for 3-4 watts in the same size beam as the dual knife-edged setup.
Unfortunately, you'd increase the vertical divergence in doing so and wind up with square divergence figures again.
Hope that helps; someone please correct me if I'm wrong.
- Kyle